Evaluating the factors affecting farmer’s choice for cropping pattern and assessing the potential of DESIS hyperspectral data to discriminate the cropping patterns

Cropping patterns refer to the annual sequence and spatial arrangement of crops on land, such as monocropping and intercropping. Acquiring information and understanding the location, extent and types of cropping patterns is essential for food security assessments. However, underlying factors that influence farmers’ choices about which cropping pattern to adopt are still unclear in most regions in the world. Technologies such as Remote sensing has enabled mapping of cropping patterns, though challenges remain in small-holder agriculture regions like Sub-Saharan Africa due to small field sizes and highly fragmented landscapes. This study first explores farmers’ knowledge through surveys to evaluate factors influencing the choice of maize cropping patterns in Busia, Kenya. Six factors are analyzed Field size, Household needs, Availability of resources, Farmers’ experience, Market demand and Pest control. Further, DESIS hyperspectral satellite data were used to characterize the monocropped and inter- cropped maize fields. Reflectance values were extracted from field boundary data and statistical tests assessed spectral differences. A Random Forest (RF) classifier was then used for feature selection, where we identified optimal spectral bands for classification. Statistical analysis revealed significant spectral differences between monocropped and intercropped maize. Key bands in the visible, red-edge, and near-infrared (NIR) regions were most effective in distinguishing cropping patterns. Feature selection prioritized five dominant bands in red-edge and NIR for classification. Using an RF classifier, the study achieved an overall accuracy (OA) of 74%, with producer accuracy (PA) of 71 % for monocropped maize and 80% for intercropped maize. User accuracy (UA) was 91 % for monocropped fields and 50% for intercropped fields. The kappa coefficient was 0.43, indicating the complexity of discriminating and classifying the maize-based cropping patterns. This study demonstrates the potential of hyperspectral remote sensing for distinguishing maize cropping patterns, though challenges exist, especially in the late maize development stage. Therefore, acquiring remote sensing images early in the growing season before the maize canopy obscures the smaller intercropped crops, is recommended. In addition, combining farmers’ perspectives with remote sensing is required for an in-depth understanding and addressing current food security challenges.